Table 1 Model’s comparative review Analysis.
Reference | Method | Main Objectives | Findings | Limitations |
|---|---|---|---|---|
ConvTrans-OptBiSVM | Hybrid convolution-transformer model for abnormal behavior detection | Improved real-time detection accuracy in crowded spaces | High computational complexity for large Scale deployment | |
Unsupervised Group Based Behavior Detection | Crowd dynamics analysis using unsupervised learning | Effective for tracking group movements in dynamic environments | Struggles with individual-level anomaly detection | |
Spatio-Temporal Saliency Descriptor & Fuzzy Representation | Anomaly detection using saliency features | Enhanced feature representation improves detection accuracy | Limited adaptability to rapidly changing environments | |
CNN Based Anomalous Behavior Detection on Buses | Detection of abnormal passenger activities | Effective detection of theft and violent behaviors in transit systems | Performance declines in highly occluded scenes | |
CNN Based Aerial Intelligence for Human Action Recognition | Drone Based surveillance for anomaly detection | Achieves high accuracy in open Space monitoring | Limited applicability in indoor or occluded settings | |
Two Stream Extreme Learning Machine & Stacked Autoencoder | Multi Stream learning for anomaly classification | Reduces dependency on large training datasets | Susceptible to overfitting on small datasets | |
MP-Abr Multi-Person Anomaly Recognition | Attention Based framework for multi-person anomaly detection | Improved robustness in multi-agent environments | Lacks adaptability to highly dynamic backgrounds | |
Efficient-X3D Based Crowd Behavior Analysis | Pre-trained network for violence detection | Achieves real-time crowd violence identification | Limited performance in low-resolution videos | |
Spatio-Temporal Autoencoders & ConvLSTMs | Autoencoder Based anomaly detection with LSTMs | Improved temporal consistency in anomaly classification | High computational cost for real-time deployment | |
Patch Based 3D Convolution with Recurrent Model | Motion Based anomaly classification | Effective in detecting anomalies with distinct motion cues | Requires large datasets for accurate training | |
Deep Learning for Abnormal Cell Division Classification | Classification of normal and abnormal cell behavior | High precision in biological anomaly detection | Limited to biomedical applications | |
Adaptive Loitering Anomaly Detection | Detecting suspicious pedestrian behaviors | Effective in monitoring unauthorized presence in restricted areas | Prone to misclassification in dense urban spaces | |
Comprehensive Review of Video Anomaly Detection Datasets | Evaluating publicly available datasets | Provides benchmark comparisons for future research | No experimental implementation | |
YOLOv8 Based Fall Detection | Human safety monitoring for assisted living | Real-time response to sudden human falls | Limited generalization to other human activities | |
Kernel Local Component Analysis & Deep Learning | Surveillance Based activity anomaly detection | High recall for behavior Based anomalies | High training overhead for complex motion patterns | |
Smart Proctoring with Automated Anomaly Detection | AI-driven proctoring for exam integrity | Improved detection of cheating behaviors in online assessments | Susceptible to false positives in ambiguous scenarios | |
Temporal Graph Attention with Transformer-Augmented RNNs | Hybrid deep learning for enhanced anomaly detection | High accuracy in complex surveillance environments | Requires extensive hyperparameter tuning | |
Vista-Lite Smart Surveillance System | Motion prediction for urban video monitoring | Reduces false alarms in high-traffic zones | High initial setup cost | |
Unsupervised Spatial–Temporal Action Translation | Unsupervised framework for violence detection | Achieves high accuracy in crime detection | Poor adaptation to new environmental conditions | |
Survey on Video Anomaly Detection | Comparative study on detection techniques | Comprehensive assessment of existing approaches | No direct performance validation | |
3D CNN with Convolutional Block Attention | Key-frame Based violence detection | Increased efficiency in event classification | Computationally expensive for real-time applications | |
Mobile Crowd Sensing in IOTA Framework | Deep learning for IoT Based anomaly detection | Efficient anomaly recognition in mobile surveillance | Scalability concerns in large urban networks | |
Suspicious Object Detection with ML | Object Based anomaly detection | Enhanced security monitoring in public places | Prone to misclassification in cluttered environments | |
SED NET: Suspicious Event Detection | Neural network Based anomaly detection | Real-time response to suspicious activities | Requires extensive training data | |
Appearance & Motion Based Crowd Anomaly Detection | Multi-object tracking for crowd event recognition | High recall in pedestrian-heavy environments | Struggles with occlusion and lighting changes | |
Psychophysiological Analysis in Emergency Events | Behavioral assessment in crisis situations | Provides insights into human reactions during accidents | Not directly applicable to real-time surveillance | |
Deep Learning Based Crowd Behavior Analysis | Crowd anomaly detection via deep feature extraction | High accuracy in public event monitoring | Limited adaptability to sudden environmental shifts | |
Enhanced YOLOv5 for Person Detection | Object recognition and motion tracking | Robust detection of individuals in varying lighting conditions | High computational cost for real-time tracking | |
RPCA-MFTSL & PSO-CNN for Anomalous Action Detection | Hybrid deep learning for motion Based anomalies | Improved anomaly generalization in surveillance videos | Performance degradation in extreme occlusions | |
AI Based Covid-19 Detection & Prevention | AI-driven medical anomaly recognition | Effective in screening Covid-19 symptoms in imaging | Limited scope beyond healthcare | |
Trajectory Based Cattle Classification | Motion feature analysis for livestock monitoring | High accuracy in predicting animal behaviors | Not applicable for human activity monitoring | |
Hierarchical Deep Dyna-Q Network (HDDQN) for action prediction, | To improve person detection, action prediction and video synthesis in CCTV surveillance. | Achieved high accuracy. Improved video synthesis with SSIM of 0.75 and PSNR of 30 dB. | High computational complexity, requiring substantial processing power | |
3D CNNs with Temporal Attention Networks for spatiotemporal action localization. | Improving person tracking, weapon detection and action recognition across multi-camera setups. | Achieved high detection accuracy. Real-time learning enhanced adaptability. | Dependency on human feedback in RL-HITL for continuous improvement. | |
DBSCAN Based Access Control | Security access validation for medical systems | Reduces false positives in authentication systems | Limited applicability in surveillance settings | |
Crowd Behavior Analysis with Expectation–Maximization | Video segmentation for behavioral classification | High precision in large Scale event analysis | Requires extensive labeled datasets | |
Autonomous Vehicle Performance Analysis | Real-road field test for self-driving systems | Evaluates driving anomalies in real-world conditions | Limited to automotive research | |
YOLOv3 & ReID Based People Flow Analysis | Statistical model for public space monitoring | Enhanced efficiency in detecting high-density pedestrian movement | Limited performance in extreme lighting conditions | |
Bioimaging & Theranostics | AI-driven medical diagnostics | High precision in cellular-level anomaly detection | Not applicable to video Based surveillance | |
HiEve: Large Scale Human-Centric Video Benchmark | Dataset for complex event detection | Provides diverse labeled scenarios for anomaly research | High storage and processing requirements | |
Deformable Convolution for Sports Recognition | Multi Scale feature extraction for action recognition | Effective in classifying sports activities | Not optimized for crowd anomaly detection | |
Attention Based Transformers for Human Activity Detection | Transformer Based activity recognition | High accuracy in event categorization | Requires high computational resources | |
Violence Detection with Mosaicking & DFE-WLSRF | Deep learning Based violence detection | Effective in classifying aggressive behaviors | Struggles with partial occlusions |
